ISSN 1364-0380 (on line) 1465-3060 (printed)
Geometry & Topology
831
G
T
G G TT TG T T
G
G T
T
G
T
G
T
G
G T GG TT
GGG T T
Volume 8 (2004) 831–876
Published: 2 June 2004
Heegaard splittings of graph manifolds
Jennifer Schultens
Department of Mathematics
1 Shields Avenue
University of California
Davis, CA 95616, USA
Email: jcs@math.ucdavis.edu
Abstract
Let M be a totally orientable graph manifold with characteristic submanifold
T and let M = V ∪S W be a Heegaard splitting. We prove that S is standard.
In particular, S is the amalgamation of strongly irreducible Heegaard splittings.
The splitting surfaces Si of these strongly irreducible Heegaard splittings have
the property that for each vertex manifold N of M , Si ∩ N is either horizontal,
pseudohorizontal, vertical or pseudovertical.
AMS Classification numbers
Primary: 57N10
Secondary: 57N25
Keywords:
Graph manifolds, Heegaard splitting, horizontal, vertical
Proposed: Cameron Gordon
Seconded: Joan Birman, Wolfgang Metzler
c Geometry & Topology Publications
Received: 30 June 2003
Revised: 1 June 2004
Jennifer Schultens
832
1
Introduction
The subject of this investigation is the structure of Heegaard splittings of graph
manifolds. This investigation continues the work begun in [20], [21], [13] and
[22]. Since the publication of those papers, new techniques have been added
to the repertoire of those interested in describing the structure of Heegaard
splittings. These include the idea of untelescoping a weakly reducible Heegaard
splitting into a generalized strongly irreducible Heegaard splitting due to M
Scharlemann and A Thompson. They also include the Rubinstein–Scharlemann
graphic, as employed by D Cooper and M Scharlemann in [6]. These insights
have not left the investigation here unaffected. We hope that their role here
is a tribute to proper affinage1 . (The structural theorem given here has been
promised for rather a long time.) A similar theorem was announced by J H
Rubinstein.
The main theorems are the following, for defintions see Sections 2, 3, 4 and 5:
Theorem 1.1 Let M be a totally orientable generalized graph manifold. If
M = V ∪S W is a strongly irreducible Heegaard splitting, then S is standard.
More specifically, S can be isotoped so that for each vertex manifold Mv of M ,
S ∩ Mv is either horizontal, pseudohorizontal, vertical or pseudovertical and
such that for each edge manifold Me , S ∩ Me is characterized by one of the
following:
(1) S ∩ Me is a collection of incompressible annuli (including spanning annuli
and possibly boundary parallel annuli) or is obtained from such a collection by
ambient 1–surgery along an arc which is isotopic into ∂Me .
(2) Me is homeomorphic to (torus) × I and there is a pair of simple closed
curves c, c′ ⊂ (torus) such that c ∩ c′ consists of a single point p ∈ (torus) and
either V ∩ ((torus) × I) or W ∩ ((torus) × I) is a collar of (c × {0}) ∪ (p × I) ∪
(c′ × {1}).
In the general case we can say the following:
Theorem 1.2 Let M be a totally orientable graph manifold. Let M = V ∪S W
be an irreducible Heegaard splitting. Let M = (V1 ∪S1 W1 ) ∪F1 (V2 ∪S2 W2 ) ∪F2
· · ·∪Fm−1 (Vm ∪Sm Wm ) be a weak reduction of M = V ∪S W . Set Mi = Vi ∪Wi .
Then Mi is a totally orientable generalized graph manifold and Mi = Vi ∪Si Wi
1
This is a French noun describing the maturing process of a cheese.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
833
is a strongly irreducible Heegaard splitting. In particular, Mi = Vi ∪Si Wi is
standard.
P
Here χ(S) = i (χ(Si ) − χ(Fi )).
Theorem 1.3 Let M be a totally orientable graph manifold. If M = V ∪S W
is an irreducible Heegaard splitting, then it is the amalgamation of standard
Heegaard splittings of generalized graph submanifolds of M .
The graph manifolds considered here are totally orientable, that is, they are
orientable 3–manifolds and for each vertex manifold the underlying surface of
the orbit space is orientable. It follows from [10, VI.34] that an incompressible
surface can be isotoped to be either horizontal or vertical in each vertex manifold of a totally orientable graph manifold. In conjunction with the notion of
untelescoping a weakly reducible Heegaard splitting into a strongly irreducible
generalized Heegaard splitting, this observation reduces the investigation at
hand to the investigation of strongly irreducible Heegaard splittings of generalized graph manifolds (for definitions, see below).
In the investigation of strongly irreducible Heegaard splittings of generalized
graph manifolds, the nice properties of strongly irreducible Heegaard splittings
often reduce this investigation to a study of the behaviour of the Heegaard
splittings near the characteristic submanifolds. In this context, a theorem of
D Cooper and M Scharlemann completes the description of this behaviour,
see Proposition 7.15 and Proposition 7.23. This theorem may be found in [6,
Theorem 4.2].
The theorem here is purely structural in the sense that it describes the various ways in which a Heegaard splitting can be constructed. Specifically, there
are finitely many possible constructions. Thus a totally orientable graph manifold possesses only finitely many Heegaard splittings up to isotopy (and hence
also up to homeomorphism). It would be possible to extract a formula for the
genera of these Heegaard splittings. However, this formula would be long, cumbersome and not very enlightening. But note that, in particular, the program
here enables a computation of Heegaard genus, ie, the smallest possible genus
of a Heegaard splitting, for totally orientable graph manifolds. To compute
this genus, one need merely consider the finitely many possible constructions,
compute the corresponding Euler characteristics, and find the extremal value.
This line of thought is pursued in [24], where the genus of a certain class of totally orientable graph manifolds is compared to the rank, ie, the least number
of generators, of the fundamental group of these manifolds.
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
834
The theorem leaves open the question of classification. There will be some,
though probably not too many, cases in which the various constructions are
isotopic. More interestingly, there may be larger scale isotopies. Ie, there may
be two Heegaard splittings of a graph manifold that are isotopic but not via an
isotopy fixing their intersection with the decomposing tori. Clearly, this leaves
much room for further investigation.
The global strategy is as follows: By Theorem 3.10, a Heegaard splitting is the
amalgamation of the strongly irreducible Heegaard splittings arising in any of
its weak reductions. Thus, one begins with a Heegaard splitting of a graph
manifold. One then considers a weak reduction of this Heegaard splitting. Cutting along the incompressible surfaces in the weak reduction yields generalized
graph manifolds with strongly irreducible Heegaard splittings. One analyzes
the possible strongly irreducible Heegaard splittings of generalized graph manifolds. Finally, one considers all possibilities arising in the amalgamation of
strongly irreducible Heegaard splittings of generalized graph manifolds.
I wish to thank the many colleagues who have reminded me that a complete
report on this investigation is past due. Among these are Ian Agol, Hugh
Howards, Yoav Moriah, Marty Scharlemann, Yo’av Rieck, Eric Sedgwick and
Richard Weidmann. I also wish to thank the MPIM-Bonn where part of this
work was done. This work was supported in part by the grant NSF-DMS
0203680.
2
Totally orientable graph manifolds
For standard definitions pertaining to knot theory see for instance [4], [11] or
[15]. For 3–manifolds see [9] or [10]. Note that the terminology for graph
manifolds has not been standardized.
Definition 2.1 A Seifert manifold is a compact 3–manifold that admits a
foliation by circles.
For a more concrete definition, see for instance [10]. The fact that the simple
definition here is in fact equivalent to more concrete definitions follows from [7].
Definition 2.2 The circles in the foliation of a Seifert fibered space M are
called fibers. The natural projection that sends each fiber to a point is denoted
by p : M → Q. The quotient space Q is called the base orbifold of M . A fiber
f is called an exceptional fiber if nearby fibers wind around f more than once.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
835
Otherwise, f is called a regular fiber. The image under p of a regular fiber is
called a regular point and the image under p of an exceptional fiber is called
an exceptional point.
The base orbifold is in fact a surface. This follows from standard facts about
foliations in conjunction with [7]. It also follows that there will be only finitely
many exceptional fibers.
Definition 2.3 For Y a submanifold of X , we denote an open regular neighborhood of Y in X by η(Y, X), or simply by η(Y ), if there is no ambiguity
concerning the ambient manifold. Similarly, we denote a closed regular neighborhood by N (Y, X), or simply by N (Y ), if there is no ambiguity concerning
the ambient manifold.
Definition 2.4 A surface S in a Seifert fibered space M is vertical if it consists
of fibers. It is horizontal if it intersects all fibers transversely. It is pseudohorizontal if there is a fiber f ⊂ M such that S ∩ (M \η(f )) is horizontal and
S ∩ N (f ) is a collar of f .
It follows that a horizontal surface in a Seifert fibered space M orbifold covers
the base orbifold of M .
Definition 2.5 A Seifert fibered space is totally orientable if it is orientable
as a 3–manifold and has an orientable base orbifold.
We are now ready to define graph manifolds.
Definition 2.6 A graph manifold is a 3–manifold M modelled on a finite
graph Γ as follows:
(1) Each vertex v of Γ corresponds to a Seifert fibered space, denoted by Mv
and called a vertex manifold;
(2) Each edge e of Γ corresponds to a 3–manifold homeomorphic to (torus) ×
S 1 , denoted by Me and called an edge manifold;
(3) If an edge e is incident to a vertex v , then this incidence is realized by an
identification of a boundary component of Me with a boundary component of
Mv via a homeomorphism.
A graph manifold is totally orientable if each vertex manifold is totally orientable.
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
836
The union of edge manifolds in M is also called the characteristic submanifold
of M . It is denoted by E . The image of a boundary component of the characteristic submanifold of M is a torus called a decomposing torus. It is denoted
by T .
Figure 1: A model graph for a graph manifold
A decomposing torus is, of course, also the image of a boundary component of
a vertex manifold. But the converse is not always true.
Remark 2.7 We have placed no restrictions on the homeomorphism that identifies a boundary component of an edge manifold with a boundary component
of a vertex manifold. Thus according to this definition, there will be Seifert
fibered spaces that admit a description as a graph manifold with non empty
characteristic submanifold. From the point of view of the investigation here,
this is often a useful way to think of such a Seifert fibered space. See [26].
Definition 2.8 A boundary component of a vertex manifold Mv of a graph
manifold M that is also a boundary component of M is called an exterior
boundary component of Mv . We denote the union of exterior boundary components of Mv by ∂E Mv .
3
Untelescoping and amalgamation
We here give the basic definitions concerning Heegaard splittings, strongly irreducible Heegaard splittings, untelescopings and amalgamations. Theorem 3.10
below is crucial to the global strategy employed in our investigation.
Definition 3.1 A compression body is a 3–manifold W obtained from a connected closed orientable surface S by attaching 2–handles to S×{0} ⊂ S×I and
capping off any resulting 2–sphere boundary components. We denote S × {1}
by ∂+ W and ∂W \∂+ W by ∂− W . Dually, a compression body is a connected
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
837
orientable 3–manifold obtained from a (not necessarily connected) closed orientable surface ∂− W × I or a 3–ball by attaching 1–handles.
In the case where ∂− W = ∅ (ie, in the case where a 3–ball was used in the dual
construction of W ), we also call W a handlebody. If W = ∂− W × I , we say
that W is a trivial compression body.
Definition 3.2 A spine of a compression body W is a 1–complex X such that
W collapses to ∂− W ∪ X .
Definition 3.3 A set of defining disks for a compression body W is a set
of disks {D1 , . . . , Dn } properly imbedded in W with ∂Di ⊂ ∂+ W for i = 1,
. . . , n such that the result of cutting W along D1 ∪ · · · ∪ Dn is homeomorphic
to ∂− W × I or to a 3–ball in the case that W is a handlebody.
Definition 3.4 A Heegaard splitting of a 3–manifold M is a decomposition
M = V ∪S W in which V , W are compression bodies, V ∩ W = ∂+ V = ∂+ W =
S and M = V ∪ W . We call S the splitting surface or Heegaard surface.
The notion of strong irreducibility of a Heegaard splitting was introduced by
A. Casson and C. McA. Gordon in [5] and has proven extremely useful.
Definition 3.5 A Heegaard splitting M = V ∪S W is strongly irreducible if
for any pair of essential disks D ⊂ V and E ⊂ W , ∂D ∩ ∂E 6= ∅.
Recall also the following related definitions:
Definition 3.6 A Heegaard splitting M = V ∪S W is reducible if there exists a
pair of essential disks D ⊂ V and E ⊂ W such that ∂D = ∂E . If M = V ∪S W
is not reducible, then it is irreducible.
A Heegaard splitting M = V ∪S W is stabilized if there exists a pair of essential
disks D ⊂ V and E ⊂ W such that |∂D ∩ ∂E| = 1.
Though all compact 3–manifolds admit Heegaard splittings, many do not admit
strongly irreducible Heegaard splitting. This fact prompted M. Scharlemann
and A. Thompson to introduce the following notion of generalized Heegaard
splittings.
Geometry & Topology, Volume 8 (2004)
838
Jennifer Schultens
Definition 3.7 A generalized Heegaard splitting of a compact orientable 3–
manifold M is a decomposition M = (V1 ∪S1 W1 ) ∪F1 (V2 ∪S2 W2 ) ∪F2 · · · ∪Fm−1
(Vm ∪Sm Wm ) such that each of the Vi and Wi is a union of compression bodies
with ∂+ Vi = Si = ∂+ Wi and ∂− Wi = Fi = ∂− Vi+1 .
We say that a generalized Heegaard splitting is strongly irreducible if each Heegaard splitting of a component of Mi = Vi ∪Si Wi is strongly irreducible and
each Fi is incompressible in M . We will denote ∪i Fi by F and ∪i Si by S .
The surfaces in F are called the thin levels and the surfaces in S the thick
levels.
Let M = V ∪S W be an irreducible Heegaard splitting. We may think of M as
being obtained from ∂− V × I by attaching all 1–handles in V (dual definition
of compression body) followed by all 2–handles in W (standard definition of
compression body), followed, perhaps, by 3–handles. An untelescoping of M =
V ∪S W is a rearrangement of the order in which the 1–handles of V and
the 2–handles of W are attached yielding a generalized Heegaard splitting.
A weak reduction of M = V ∪S W is a strongly irreducible untelescoping of
M = V ∪S W .
Note that a weak reduction of a strongly irreducible Heegaard splitting would
just be the strongly irreducible Heegaard splitting itself. The Main Theorem
in [18] implies the following:
Theorem 3.8 Let M be an irreducible 3–manifold. Any Heegaard splitting
M = V ∪S W has a weak reduction.
Definition 3.9 Let N, L be 3–manifolds with R a closed subsurface of ∂N ,
and S a closed subsurface of ∂L, such that R is homeomorphic to S via a
homeomorphism h. Further, let (U1 , U2 ), (V1 , V2 ) be Heegaard splittings of
N, L such that N (R) ⊂ U1 , N (S) ⊂ V1 . Then, for some R′ ⊂ ∂N \R and
S ′ ⊂ ∂L\S , U1 = N (R∪R′ )∪(1−handles) and V1 = N (S ∪S ′ )∪(1−handles).
Here N (R) is homeomorphic to R × I via a homeomorphism f and N (S) is
homeomorphic to S × I via a homeomorphism g . Let ∼ be the equivalence
relation on N ∪ L generated by
(1) x ∼ y if x, yǫη(R) and p1 · f (x) = p1 · f (y),
(2) x ∼ y if x, yǫη(S) and p1 · g(x) = p1 · g(y),
(3) x ∼ y if xǫR, yǫS and h(x) = y ,
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
839
where p1 is projection onto the first coordinate. Perform isotopies so that
for D an attaching disk for a 1–handle in U1 , D′ an attaching disk for a 1–
handle in V1 , [D] ∩ [D ′ ] = ∅. Set M = (N ∪ L)/ ∼, W1 = (U1 ∪ V2 )/ ∼,
and W2 = (U2 ∪ V1 )/ ∼. In particular, (N (R) ∪ N (S)/ ∼) ∼
= R, S . Then
′
W1 = V2 ∪ N (R ) ∪ (1 − −handles), where the 1–handles are attached to
∂+ V2 and connect ∂N (R′ ) to ∂+ V2 , and hence W1 is a compression body.
Analogously, W2 is a compression body. So (W1 , W2 ) is a Heegaard splitting of
M . The splitting (W1 , W2 ) is called the amalgamation of (U1 , U2 ) and (V1 , V2 )
along R, S via h.
Theorem 3.8 together with [20, Proposition 2.8] implies the following:
Theorem 3.10 Suppose M = V ∪S W is an irreducible Heegaard splitting
and M = (V1 ∪S1 W1 ) ∪F1 (V2 ∪S2 W2 ) ∪F2 · · · ∪Fm−1 (Vm ∪Sm Wm ) a weak
reduction of M = V ∪S W . Then the amalgamation of M = (V1 ∪S1 W1 ) ∪F1
(V2 ∪S2 W2 ) ∪F2 · · · ∪Fm−1 (Vm ∪Sm Wm ) along F1 ∪ · · · ∪ Fm−1 is M = V ∪S W .
One of the nice properties of strongly irreducible Heegaard splittings is apparent
in the following lemma which is a deep fact and is proven, for instance, in [23,
Lemma 6].
Lemma 3.11 Suppose M = V ∪S W is a strongly irreducible Heegaard splitting and P ⊂ M an essential incompressible surface. Then S can be isotoped
so that S ∩ P consists only of curves essential in both S and P .
4
Incompressible surfaces and generalized graph
manifolds
In the arguments that follow, we employ the ideas of untelescoping and amalgamation. In this section, we describe the incompressible surfaces that arise in
a weak reduction of a Heegaard splitting. We then describe the 3–manifolds
that result from cutting a graph manifold along such incompressible surfaces.
We will call these 3–manifolds generalized graph manifolds. Later, we will consider the strongly irreducible Heegaard splittings on these generalized graph
manifolds.
Remark 4.1 An edge manifold of a totally orientable graph manifold M is
homeomorphic to (torus) × I . There are infinitely many distinct foliations of
Geometry & Topology, Volume 8 (2004)
840
Jennifer Schultens
(torus) × I as an annulus bundle over the circle. The incompressible surfaces in
(torus) × I are tori isotopic to (torus) × {point}, annuli isotopic to the annular
fibers in the foliations of (torus) × I as an annulus bundle over the circle and
annuli parallel into ∂((torus) × I).
Lemma 4.2 Let F be an incompressible surface in a totally orientable graph
manifold M . Then F may be isotoped so that in each edge manifold it consists
of incompressible tori and essential annuli and in each vertex manifold it is
either horizontal or vertical.
Proof Let T be the collection of decomposing tori for M . Since F and T
are incompressible, F may be isotoped so that F ∩ T consists only of curves
essential in both F and T . We may assume that this has been done in such a
way that the number of components in F ∩T is minimal. Let N be a component
of M \T , then F ∩ N is incompressible. Furthermore, no component of F ∩ N
is an annulus parallel into T .
Suppose F ∩ N is boundary compressible in N . Let D̂ be a boundary compressing disk for F ∩ N . Then ∂ D̂ = a ∪ b, with a ⊂ ∂N and b ⊂ F .
Since F ∩ T consists only of curves essential in both F and T , the component A of ∂N \(F ∩ ∂N ) that contains a is an annulus. Let B(D̂) be a
bicollar of D̂ . Then ∂B(D̂) has two components, D̂0 , D̂1 . Consider the disk
D = (A\(A ∩ B(D̂)) ∪ D̂0 ∪ D̂1 . Since F is incompressible, D must be parallel
to a disk in F , but this implies that the number of components of F ∩ T is not
minimal, a contradiction. Thus, F ∩ N is boundary incompressible in N .
If N is an edge manifold, then F ∩ N is as required by Remark 4.1. If N is a
vertex manifold, then [10, VI.34] allows three possibilities for F ∩ N : (1) F ∩ N
is vertical; (2) F ∩ N is horizontal; or (3) F ∩ N is the boundary of a twisted
I –bundle over a horizontal surface F̂ N in N . For a boundary incompressible
surface in N this latter possibility would imply that there is a nonorientable
horizontal surface F̂ N in C . In particular, F̂ N would be a cover of the base
orbifold. But this is impossible. Hence F ∩ N is either horizontal or vertical.
Hence F ∩ N is as required.
The following definition describes the 3–manifolds that result when a totally
orientable graph manifold is cut along incompressible surfaces.
Definition 4.3 A generalized graph manifold is a 3–manifold M modelled on
a finite graph Γ as follows:
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
841
(1) Each vertex v of Γ corresponds either to a Seifert fibered space or to a 3–
manifold homeomorphic to (compact surface) × [0, 1]. This manifold is denoted
by Mv and called a vertex manifold.
(2) Each edge e of Γ corresponds either to a 3–manifold homeomorphic to
(torus) × I or to a 3–manifold homeomorphic to (annulus) × I . This manifold
is denoted by Me and called an edge manifold.
(3) If the edge manifold Me is homeomorphic to (torus) × I and e is incident
to a vertex v , then this incidence is realized by an identification of a boundary
component of Me with a boundary component of Mv . In particular, Mv must
be Seifert fibered.
(4) If the edge manifold Me is homeomorphic to (annulus) × I and e is incident to a vertex v , then this incidence is realized by an identification of a
component of (∂(annulus)) × I , with a subannulus of ∂Mv . If Mv is Seifert
fibered, then this subannulus of ∂Mv consists of fibers of Mv . If Mv is homeomorphic to (compact surface) × I , then this subannulus is a component of
(∂(compact surface)) × I .
(5) If a vertex manifold Mv is homeomorphic to (compact surface) × I , then
the valence of v equals the number of components of (∂(compact surface)) × I .
Ie, each component of (∂(compact surface)) × I is identified with a subannulus
of the boundary of an edge manifold.
A generalized graph manifold is totally orientable if each vertex manifold that is
Seifert fibered is totally orientable and each vertex manifold that is not Seifert
fibered is homeomorphic to (compact orientable surface) × I .
The union of edge manifolds in M is also called the characteristic submanifold
of M . It is denoted by E . The image of a torus or annulus, respectively, along
which an indentification took place is called a decomposing torus or decomposing annulus, respectively. The union of decomposing tori and annuli is denoted
by T .
Consider the case in which an edge manifold Me = (torus) × [0, 1] of a graph
manifold is cut along an incompressible torus T = (torus)×{point}. When Me
is cut along T , the remnants of Me are (torus)×[0, 12 ] and (torus)×[ 21 , 1]. Each
of these remnants forms a collar of a vertex manifold. A foliation of (torus)
by circles may be chosen in such a way that the Seifert fibration of the vertex
manifold extends across the remnant. We may thus ignore these remnants, as
we do in the above definition of generalized graph manifolds. This facilitates
the discussion of strongly irreducible Heegaard splittings of generalized graph
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
842
manifolds. Later, when considering amalgamations of strongly irreducible Heegaard splittings of generalized graph manifolds, we will have to reconsider these
remnants.
Definition 4.4 A portion of the boundary of a vertex manifold Mv of a generalized graph manifold M that is contained in ∂M is called an exterior boundary
component of Mv . We denote the union of exterior boundary components of
Mv by ∂E Mv .
5
Motivational examples of Heegaard splittings
In this section we describe some examples of Heegaard splittings for graph
manifolds and generalized graph manifolds. The following definition facilitates
describing the structure of certain surfaces.
Definition 5.1 Let F be a surface in a 3–manifold M and α an arc with
interior in M \F and endpoints on F . Let C(α) be a collar of α in M . The
boundary of C(α) consists of an annulus A together with two disks D1 , D2 ,
which we may assume to lie in F . We call the process of replacing F by
(F \(D1 ∪ D2 )) ∪ A performing ambient 1–surgery on F along α.
The process of ambient 1–surgery on a surface along an arc is sometimes informally referred to as “attaching a tube”.
Example 5.2 Let Q be a closed orientable surface. The standard Heegaard
splitting of Q × S1 may be constructed in more than one way. In particular,
consider a small disk D ⊂ Q and a collection Γ of arcs that cut Q\D into
a disk. Let S be the result of performing ambient 1–surgery on ∂D × S1
along Γ × {point}. Then S is the splitting surface of a Heegaard splitting
Q × S1 = V ∪S W (for details, [20]). See Figure 2. One of the handlebodies is
((shaded disk) × S1 ) ∪ N (dashed arcs).
The same Heegaard splitting of Q × S1 may be obtained in another way: Partition S1 into two intervals I1 , I2 that meet in their endpoints. Consider two
distinct points p, q ∈ Q. Then the surface obtained by performing ambient
1–surgery on Q × (I1 ∩ I2 ) along (p × I1 ) ∪ (q × I2 ) is isotopic to S above.
The fact that this Heegaard splitting can be constructed either from a vertical
torus or from horizontal surfaces via ambient 1–surgery is likely to be a very
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
843
Figure 2: Schematic for Heegaard splitting of Q × S1
special feature. But there are no general techniques for detecting the sort of
global isotopies that allow this to happen.
Note also that this Heegaard splitting is not strongly irreducible. The two descriptions of this Heegaard splitting hint at distinct weak reductions. In one
of these weak reductions, the incompressible surfaces would consist of vertical
tori. In the other, the incompressible surfaces would consist of horizontal incompressible surfaces. In weak reductions of the latter type, we see that the
Heegaard splitting is an amalgamation of two Heegaard splittings with mostly
horizontal splitting surface.
The Heegaard splitting described turns out to be the only irreducible Heegaard
splitting for a manifold of the form (closed orientable surface) × S1 . This
fact is the main theorem of [20]. The first description of the construction can
be generalized to Seifert fibered spaces to provide the canonical Heegaard splittings for totally orientable Seifert fibered spaces, see [1] and [13]. The Heegaard
splittings arising from this construction have been termed vertical. This terminology has created some confusion, because the splitting surface of a vertical
Heegaard splitting is not vertical as a surface. Here we will continually focus on
the splitting surface. In particular, we will want to distinguish between surfaces
that are vertical and surfaces that are the splitting surface of a vertical Heegaard splitting. For this reason, we will augment the existing terminology and
refer to the splitting surface of a vertical Heegaard splitting as “pseudovertical”.
We recall the definition of a vertical Heegaard splitting for a Seifert fibered
space. The structure of Heegaard splittings for totally orientable Seifert fibered
spaces has been completely described in [13]. Thus we may restrict our attention
to Seifert fibered spaces with non empty boundary in the definition below.
Definition 5.3 Let M be a Seifert fibered space with ∂M 6= ∅. Denote the
base orbifold of M by O. Denote the exceptional fibers of M by f1 , . . . , fn and
the corresponding exceptional points in O by e1 , . . . , en . Denote the boundary
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
844
b2
b1
e3
e1
e2
Figure 3: Schematic for a vertical Heegaard splitting of a Seifert fibered space
components of M by B1 , . . . , Bm and the corresponding boundary components
of O by b1 , . . . , bm ,
Partition f1 , . . . , fn into two subsets: f1 , . . . , fi , the fibers that will lie in V
and fi+1 , . . . , fn , the fibers that will lie in W . Then partition B1 , . . . , Bm
into two subsets: B1 , . . . , Bj , the boundary components that will lie in V and
Bj+1 , . . . , Bm , the boundary components that will lie in W .
We may assume that j ≥ 1, for otherwise we may interchange the roles of V
and W in the construction below. Let Γ be a collection of arcs in O each
with at least one endpoint on b1 such that O\Γ is a regular neighborhood of
ei+1 ∪ · · · ∪ en ∪ bj+1 ∪ · · · ∪ bm or of a point, if this set is empty. See Figure 3.
Set V = N (f1 ∪ · · · ∪ fi ∪ B1 ∪ · · · ∪ Bj ∪ Γ) and set W = closure(M \V ). Set
S = ∂+ V = ∂+ W . Then M = V ∪S W is a Heegaard splitting. (For details,
see [21] or [13].) A Heegaard splitting of a Seifert fibered space with non empty
boundary constructed in this manner is called a vertical Heegaard splitting.
A pseudovertical surface is a surface that is the splitting surface of a vertical
Heegaard splitting.
A little more work is required to extend this notion to the setting of graph manifolds. Here we consider the intersection of a Heegaard splitting of a generalized
graph manifold with a vertex manifold that is a Seifert fibered space. Denote
the Heegaard splitting by M = V ∪S W and the vertex manifold by Mv . Here
S ∩ Mv is not necessarily connected. Furthemore, S ∩ Mv has boundary. A
concrete description of all possible such surfaces would be quite extensive. For
this reason, we give the following, purely structural, definition:
Definition 5.4 Let M = V ∪S W be a Heegaard splitting of a generalized
graph manifold with non empty characteristic submanifold. Let Mv be a Seifert
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
845
fibered vertex manifold of M . We say that S ∩ Mv is pseudovertical if the
following holds:
There is a collection of vertical annuli and tori A ⊂ Mv . And there is a
collection of arcs Γ in the interior of Mv such that each endpoint of each arc in
Γ lies in A and such that Γ projects to a collection of disjoint imbedded arcs
in Ov . And S ∩ Mv is obtained from A by ambient 1–surgery along Γ.
The assumption that M = V ∪S W is a Heegaard splitting places strong restrictions on S . If S ∩ Mv is pseudovertical, then S ∩ (∂Mv \∂E Mv ) consists
of vertical curves. Thus V ∩ (∂Mv \∂E Mv ) and W ∩ (∂Mv \∂E Mv ) consist of
annuli. Each such annulus is either a spanning annulus in V or W , or it has
both boundary components in S .
Consider the result of cutting a compression body V along an annulus A with
∂A ⊂ ∂+ V . If the annulus is inessential, then the effect is nil. If the annulus is
essential, then the result is again a, possibly disconnected, compression body,
see [23, Lemma 2]. If the annulus is a spanning annulus, then the result is
a handlebody. But in this context we should think of it as a compact 3–
manifold of the form ((compact surface) × I) ∪ (1 − handles). The only way
this can happen, given the structure of S ∩ Mv , is if the components of V ∩ Mv
and W ∩ Mv are constructed from vertical solid tori and perhaps components
homeomorphic to (annulus) × S1 by attaching “horizontal” 1–handles. For
more concrete computations, see [24].
It is a non trivial fact that for Seifert fibered spaces the two definitions of
pseudovertical surfaces coincide. This follows from [20, Proposition 2.10] via
Lemma 7.1 (an adaptation of the central argument in [21]) along with Lemma
7.4. For an illustration, see the final remarks in the example below.
Example 5.5 Let M be a Seifert fibered space with base orbifold a disk and
with two exceptional fibers f1 , f2 . Let T be a boundary parallel torus and
let α be an arc connecting T to itself that projects to an imbedded arc that
separates the two exceptional points. Let S be the result of performing ambient
1–surgery on T along α. Then S is the splitting surface of a Heegaard splitting
of M = V ∪S W (see [13]).
Now consider two copies M1 , M2 of M with Heegaard splittings Mi = Vi ∪Si Wi .
We may assume that ∂M1 ⊂ V1 and ∂M2 ⊂ W2 . We identify ∂M1 and ∂M2
to obtain a 3–manifold M̃ . Juxtaposing the two Heegaard splittings provides a
generalized strongly irreducible Heegaard splitting. It is indicated schematically
in Figure 4. Amalgamating the two Heegaard splittings along ∂M1 , ∂M2 results
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
846
in a Heegaard splitting M̃ = V ∪S W . The result is schematically indicated
in Figure 5. The circles correspond to vertical tori. The splitting surfaces of
the Heegaard splittings are obtained by performing ambient 1–surgery on these
tori along arcs in M̃ corresponding to the dashed arcs.
Figure 4: Schematic for a generalized strongly irreducible Heegaard splitting
Figure 5: Schematic for Heegaard splitting after amalgamation
The manifold M̃ obtained when ∂M1 and ∂M2 are identified is a graph manifold modelled on a graph with two vertices and one edge connecting the two
vertices. The vertex manifolds are slightly shrunken versions of M1 and M2 .
The edge manifold is a collar of the image of ∂M1 and ∂M2 in M̃ .
If the homeomorphism that identifies ∂M1 with ∂M2 is fiberpreserving, then
M̃ is in fact a Seifert fibered space. In this case S is isotopic to the surface
indicated schematically in Figure 6. This surface is obtained by performing
ambient 1–surgery along two arcs corresponding to the dashed arcs on the two
vertical tori corresponding to the two solid circles.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
847
Figure 6: Schematic for Heegaard splitting of a Seifert fibered space
The above construction can be generalized to arbitrary graph manifolds. The
resulting Heegaard splittings can be considered the canonical Heegaard splittings and generalized Heegaard splittings of graph manifolds. Contemplation of
Figure 5 might lead one to believe that amalgamations of strongly irreducible
Heegaard splittings still have enough structure to be described in the terminology used here. But this is not the case, as we shall see in the following
example.
Example 5.6 Let M1 be as above. Let P be a thrice punctured S2 . Denote
the boundary components of P by b1 , b2 , b3 . Let M2 be the 3–manifold obtained
from P × S1 by identifying b2 × S1 and b3 × S1 via a homeomorphism that takes
b2 × {point} to {point} × S1 .
Here M2 is a graph manifold modelled on a graph with one vertex and one edge.
It has a Heegaard splitting depicted schematically in Figure 7. In P × S1 , we
take the products of the regions pictured. This yields a white (annulus) × S1
and a shaded solid torus. When b2 × S1 and b3 × S1 are identified, both
(annulus) × S1 and the solid torus meet themselves in (square) disks. Thus we
obtain a (strongly irreducible) Heegaard splitting of genus 2 for M2 . Denote
the compression body by V2 and the handlebody by W2 .
This Heegaard splitting can also be constructed by ambient 1–surgery on a
boundary parallel torus along an arc as pictured in Figure 8.
The description indicated schematically in Figure 7 is preferable to the one
indicated schematically in Figure 8. This is because the splitting surface of the
former is vertical in the Seifert fibered vertex manifold and has a very special
structure in the edge manifold. The latter intersects the edge manifold in a
tube.
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
848
Figure 7: Schematic for Heegaard splitting of a graph manifold
Figure 8: Schematic for Heegaard splitting of a graph manifold
Now let M be a 3–manifold obtained by identifying the boundary component of
M1 to the boundary component of M2 . Then M is a graph manifold modelled
on a graph as in Figure 9.
Figure 9: The graph on which M is modelled
Here M inherits a strongly irreducible generalized Heegaard splitting that is
the juxtaposition of the strongly irreducible Heegaard splittings of M1 and
M2 . If we wish to amalgamate the two Heegaard splittings, recall that the
collar neighborhood of ∂M2 that lies in V2 will be identified to a single torus.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
849
To prevent this identification from interfering with the decomposing tori of
M2 , we are forced to consider the second description of the Heegaard splitting
M2 = V2 ∪S2 W2 . But this means that our description of the resulting Heegaard
splitting involves a tube which runs through an edge manifold.
The tube in the above description of the Heegaard splitting does not fit nicely
into the characterization of Heegaard splittings as horizontal, pseudohorizontal,
vertical or pseudovertical. The moral is, that such tubes arise and can be quite
complicated. A more thorough investigation of the nature of such tubes will
be the subject of a further investigation. However, such tubes do not arise
in strongly irreducible Heegaard splittings. For this reason we prefer to think
of our Heegaard splittings as amalgamations of strongly irreducible Heegaard
splittings.
The following two examples illustrate more peculiar Heegaard splittings that
arise under special circumstances.
Example 5.7 Let N be a Seifert fibered space with base orbifold the sphere
and with four exceptional fibers f1 , . . . , f4 with carefully chosen invariants:
l
1 1 1
2 , 2 , 2 , 2l+1 . Here N \η(f4 ) is a Seifert fibered manifold with boundary and
hence fibers over the circle. More specifically, it fibers as a once punctured
torus bundle over the circle. By partitioning the circle into two intervals I1 , I2
that meet in their endpoints, we obtain a decomposition N \η(f4 ) = V ′ ∪S ′ W ′
with V ′ = (once punctured torus) × I1 , W ′ = (once punctured torus) × I2 and
S ′ = two once punctured tori.
Note that this decomposition is not a Heegaard splitting in the sense used here
as S ′ is not closed. Now the carefully chosen invariants guarantee that the
boundary of a meridian disk of N (f4 ) meets ∂V ′ in a single arc. In particular,
V = V ′ ∪ N (f4 ) is also a handlebody (of genus two). Setting W = W ′ , this
defines a Heegaard splitting N = V ∪S W . The splitting surface of this Heegaard
splitting is a horizontal surface away from N (f4 ). And after a small isotopy,
S ∩ N (f4 ) is a collar of f4 . Thus S is pseudohorizontal.
Example 5.8 Let Q be a once punctured torus. Set Mi = Q × S1 for i = 1, 2.
Partition S1 into two intervals I1 , I2 meeting in their endpoints. Set Vi =
Q×I1 ⊂ Mi and Wi = Q×I2 ⊂ Mi . Let Ti = ∂Mi and ci = ∂Q×{point} ⊂ Ti .
Identify T1 and T2 via a homeomorphism so that |c1 ∩ c2 | = 1. Then V1 and
V2 meet in a (square) disk, hence V = V1 ∪ V2 is a handlebody. Similarly,
W = W1 ∪ W2 is a handlebody. Set S = ∂V = ∂W , then M = V ∪S W is a
Heegaard splitting of M = M1 ∪T1 =T2 M2 .
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
850
Here M is a graph manifold modelled on a graph with two vertices and one
edge connecting the two vertices. Its characteristic submanifold is a collar of
T1 = T2 . In the vertex manifolds, S is horizontal. In the edge manifold, S has
a very specific structure.
Definition 5.9 Let M be a generalized graph manifold with characteristic
submanifold E . Let M = V ∪S W be a Heegaard splitting. We say that M =
V ∪S W is standard if S can be isotoped so that for each vertex manifold Mv
of M , S ∩ Mv is either horizontal, pseudohorizontal, vertical or pseudovertical
and such that for each edge manifold Me of M , S ∩ Me is characterized by one
of the following:
(1) S ∩ Me is a collection of incompressible annuli (including spanning annuli
and possibly boundary parallel annuli) or is obtained from such a collection by
ambient 1–surgery along an arc which is isotopic into ∂Me .
(2) Me is homeomorphic to (torus) × I and there is a pair of simple closed
curves c, c′ ⊂ (torus) such that c ∩ c′ consists of a single point p ∈ (torus) and
either V ∩ ((torus) × I) or W ∩ ((torus) × I) is a collar of (c × {0}) ∪ (p × I) ∪
(c′ × {1}).
6
The active component
In this section we consider a generalized graph manifold W . We show that
if M = V ∪S W is a strongly irreducible Heegaard splitting, then S may be
isotoped so that it is incompressible away from a single vertex or edge manifold
of M .
Lemma 6.1 Let M be a generalized graph manifold with characteristic submanifold T . Let M = V ∪S W be a strongly irreducible Heegaard splitting. Let
DV be a collection of defining disks for V and DW a collection of defining disks
for W . There is a vertex or edge manifold N of M so that, after isotopy, each
outermost disk component of both DV \(T ∩ D V ) and of DW \(T ∩ DW ) lies in
N . Moreover, for each vertex or edge manifold Ñ 6= N , S ∩ Ñ is incompressible.
Proof Isotope S so that T ∩S consists only of curves essential in both T and
S . Furthermore, assume that the isotopy has been chosen so that |T ∩S| is
minimal subject to this condition. Suppose that D ′ is an outermost subdisk of
DV \(T ∩ D). Let N be the vertex or edge manifold of M containing D ′ . Then
either D ′ is a disk in the interior of V or D ′ meets an annular component A
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
851
of V ∩ ∂N . In case of the latter, ∂D ′ meets A in a single arc, a. Let D ′′ be
the disk obtained by cutting A along a, adding two copies of D ′ and isotoping
the result to be a properly imbedded disk in V . The assumption that |T ∩S|
be minimal guarantees that D ′′ is an essential disk in V . Thus in both cases,
there is an essential disk properly imbedded in V that lies in the interior of N ,
we refer to this disk as D.
Similarly, consider an outermost subdisk of DW \(T ∩DW ). The above argument
shows that in a vertex or edge manifold N ′ of M , there is an essential disk E
properly imbedded in W . Since M = V ∪S W is strongly irreducible, D must
meet E , hence the vertex or edge manifold N ′ must coincide with the vertex
or edge manifold N .
It follows that for each vertex or edge manifold Ñ 6= N , S ∩ Ñ is incompressible.
Definition 6.2 If M is a generalized graph manifold, with a strongly irreducible Heegaard splitting M = V ∪S W , then the vertex or edge manifold N
as in Lemma 6.1 is called the active component of M = V ∪S W .
7
What happens in the active component?
The possibilities for the active component depend on the type of the active
component. There are five possibilities. We discuss each in turn.
7.1
Seifert fibered vertex manifold with exterior boundary
We first consider the case in which the active component of M = V ∪S W
is a vertex manifold Mv that is a Seifert fibered space and that has exterior
boundary. This situation has been studied extensively in the more restricted
case in which M itself is a Seifert fibered space.
Lemma 7.1 Suppose M is a totally orientable Seifert fibered space with non
empty boundary and M = V ∪S W is a Heegaard splitting. Then each exceptional fiber of Mv is a core of either V or W .
Proof This is [21, Lemma 4.1], the central argument in [21].
Geometry & Topology, Volume 8 (2004)
852
Jennifer Schultens
Theorem 7.2 Suppose M is a totally orientable Seifert fibered space with
non empty boundary. Suppose M = V ∪S W is a Heegaard splitting. Then S
is pseudovertical.
Proof This is the main theorem of [21].
The argument extends to a more general setting. Indeed, the isotopy performed
in [21, Lemma 4.1] takes place within a small regular neighborhood of a saturated annulus. In particular, the isotopy can here be performed entirely within
Mv . The only requirements on this saturated annulus are that one boundary
component has to lie on the splitting surface of the Heegaard splitting and the
other has to wrap around the exceptional fiber.
Lemma 7.3 Suppose that M is a totally orientable graph manifold and M =
V ∪S W is a Heegaard splitting. Suppose further that there is an exceptional
fiber f in Mv and an annulus A such that:
(1) One component of ∂A wraps at least twice around f ; and
(2) A is embedded away from ∂A ∩ f ; and
(3) ∂A\f lies in S .
Then f is a core of either V or W .
Proof In fact, in the central argument in [21], the existence of a boundary
component is used exclusively to produce such an annulus.
This more general lemma will be used in the next subsection. A consequence
of Lemma 7.1 is that we can make use of the following lemma.
Lemma 7.4 Suppose that f is an exceptional fiber of Mv and that f is
also a core of V . Then M \η(f ) = (V \η(f )) ∪S W is a Heegaard splitting.
Furthermore, S ∩ (Mv \η(f )) is vertical or pseudovertical, respectively, if and
only if S ∩ Mv is vertical or pseudovertical, respectively. The same holds if f
is a core of W .
Proof Since f is a core of V , V \η(f ) is still a compression body. Thus
M \η(f ) = (V \η(f )) ∪S W is a Heegaard splitting. Conversely, if M \η(f ) =
(V \η(f )) ∪S W is a Heegaard splitting, then (V \η(f )) ∪ N (f ) is a compression
body. Hence M = V ∪S W is a Heegaard splitting.
Now by the definition of vertical and pseudovertical, respectively, S ∩(Mv \η(f ))
is vertical or pseudovertical, respectively, if and only if S ∩ Mv is vertical or
pseudovertical, respectively. Compare Figures 6 and 10.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
853
Figure 10: Schematic for vertical Heegaard splittings of graph manifold with boundary
The following Proposition generalizes Theorem 7.2. But because Theorem 7.2
is already known, we restrict our attention to the case in which M has a non
empty characteristic submanifold.
Proposition 7.5 Suppose that M is a connected generalized graph manifold
with non empty characteristic submanifold. Suppose that M = V ∪S W is a
strongly irreducible Heegaard splitting. Suppose that the active component of
M = V ∪S W is a vertex manifold Mv that is a Seifert fibered space and that
has exterior boundary. Then we may isotope the decomposing tori, thereby
redefining Mv and the edge manifolds for which e is incident to v slightly, so
that after this isotopy, S ∩ Mv is a vertical surface and so that an edge manifold
becomes the active component.
Proof The proof is by induction on the number of exceptional fibers in Mv .
Suppose first that Mv contains no exceptional fibers. Denote the exterior
boundary of Mv by ∂E Mv . Let A be a collection of disjoint essential vertical annuli in Mv that cut Mv into a regular neighborhood of ∂Mv \∂E Mv .
After an isotopy, S ∩ A consists of closed curves essential in both S and A
and in the minimal possible number of such curves. In particular, after a small
isotopy, this intersection consists of regular fibers of Mv .
Now isotope S so that S ∩ N (A) consists of vertical incompressible annuli.
Set M̃v = N (∂E Mv ) ∪ N (A). Isotope S so that S ∩ M̃v consists of S ∩ N (A)
together with annuli in N (∂E Mv )\N (A) that join two components of S ∩N (A).
Isotope any annuli in S ∩ (Mv \M̃v ) that are parallel into ∂ M̃v into M̃v . Set
∂E M̃v = ∂E Mv .
Let T̃ be a component of ∂ M̃v \∂E M̃v . Then T̃ is parallel to a decomposing
torus or annulus T . We replace T by T̃ . We may do so via an isotopy. We do
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
854
Figure 11: The active component is a vertex manifold
Figure 12: The active component is an edge manifold
this for all components of ∂ M̃v \∂E M̃v . After this process, the conclusions of
the proposition hold. See Figures 11 and 12.
To prove the inductive step, suppose that f is an exceptional fiber of Mv .
Then by Lemma 7.1, f is a core of either V or W , say of V . The inductive
hypothesis in conjunction with Lemma 7.4 then proves the theorem.
7.2
Seifert fibered vertex manifold without exterior boundary
Next we consider the case in which the active component of M = V ∪S W is
a vertex manifold Mv that is a Seifert fibered space and that has no exterior
boundary. Part of the strategy is somewhat reminiscent of the strategy used in
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
855
the preceding case. However, the situation here is more complicated and a more
refined strategy must be used. The refined strategy involves a generalization of
Lemma 3.11 to a “spine” of the vertex manifold.
Definition 7.6 Let Qv be the base orbifold of Mv . A spine for Qv is a 1–
complex Γ1 with exactly one vertex v that cuts Qv into a regular neighborhood
of ∂Qv ∪ (exceptional points). A 2–complex of the form Γ2 = p−1 (Γ1 ) is called
a spine of Mv . We denote the regular fiber p−1 (v) by γ .
The following lemma generalizes Lemma 3.11. Though we will only be interested in this lemma in the case that N is a vertex manifold of a graph manifold,
we state it in very general terms. It applies to a larger class of 3–manifolds than
just graph manifolds.
Lemma 7.7 Let M = V ∪S W be a strongly irreducible Heegaard splitting.
Let N be a totally orientable Seifert fibered submanifold of M that doesn’t
meet ∂M . Let Γ2 be a spine of N . Then S may be isotoped so that the
following hold:
(1) S ∩ Γ2 consists of simple closed curves and simple closed curves wedged
together at points in γ .
(2) No closed curve in S ∩ Γ2 bounds a disk in Γ2 \S .
Proof The first part of the assertion follows by general position. To prove
the second assertion, let X be a spine of V and Y a spine of W . Then
M \(∂− V ∪ X ∪ ∂− W ∪ Y ) is homeomorphic to S × (0, 1). X can’t be disjoint
from Γ2 . Thus for t near 0, (S × t) ∩ Γ2 contains simple closed curves that
bound essential disks in V . Y can’t be disjoint from Γ2 either. Thus for t near
1, (S × t) ∩ Γ2 contains simple closed curves that bound essential disks in W .
As t increases, (S ×t)∩Γ2 changes continuously. Since M = V ∪S W is strongly
irreducible, there can be no t such that (S ×t)∩Γ2 contains simple closed curves
that bound essential disks in V and simple closed curves that bound essential
disks in W . Thus, there is a t0 , such that (S ×t0 )∩Γ2 contains no simple closed
curves that bound essential disks in V or W . Any remaining disk components
in (S × t0 ) ∩ Γ2 must be inessential in V or W and can hence be removed via
isotopy. The lemma follows.
Before launching into the two main portions of the argument, we prove an
auxiliary lemma. This lemma is a weak version of a counterpart to Theorem
3.3 in [17].
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
856
Figure 13: Schematic for a Heegaard splitting intersecting a solid torus
Lemma 7.8 Suppose M = V ∪S W is a strongly irreducible Heegaard splitting
of a 3–manifold M . Suppose U ⊂ M is a solid torus such that S intersects
∂U in meridians. Further suppose that ∂(M \interior(U )) is incompressible in
M \interior(U ). Then S ∩ U consists of meridian disks of U and components
that are obtained by ambient 1–surgery on pairs of meridian disks along a single
arc that joins the two meridian disks.
Proof Let s be a component of S ∩ ∂U . A collar of s in S is an annulus
A. Lemma 2.6 in [19] states: “Suppose S gives a Heegaard splitting of a 3–
manifold M into compression bodies V and W . Suppose that F ⊂ S is a
compact subsurface so that every component of ∂F is essential in S . Suppose
each component of ∂F bounds a disk in M disjoint from interior(F ). Either
∂F bounds a collection of disks in a single compression body or M = V ∪S W
is weakly reducible.”
Here M = V ∪S W is strongly irreducible. It follows that either s bounds
a disk in S or s bounds a disk in a single compression body, say V . Here
∂(M \interior(U )) is incompressible in M \interior(U ). It follows that in case
of the former, s bounds a meridian disk of U in S ∩ U and that in case of the
latter, a meridian disk of U bounded by s lies entirely in V .
Thus either s bounds a meridian disk in S ∩ U or bounds a meridian disk in
the surface obtained by compressing S ∩ U along D. It follows that S ∩ U may
be reconstructed from meridian disks by ambient 1–surgery along a collection
of arcs such that each meridian disk meets at most one endpoint of one of the
arcs. In particular, between two meridian disks there is at most one arc.
Lemma 7.9 Suppose that the active component of M = V ∪S W is a vertex
manifold Mv that is a Seifert fibered space with boundary but no exterior
boundary. Then one of the following holds:
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
857
(1) We may isotope the decomposing tori, thereby redefining Mv and the edge
manifolds for which e is incident to v slightly, so that after this isotopy, S ∩ Mv
is either a horizontal or vertical incompressible surface and so that an edge
manifold becomes the active component.
Or:
(2) S may be isotoped within Mv so that a fiber f of Mv lies in S .
Proof Let Γ2 be a spine of Mv and isotope S within Mv so that the conclusions of Lemma 7.7 hold. Three cases need to be considered:
Case 1 A simple closed curve in S ∩ Γ2 can be isotoped to be vertical.
Then S may be isotoped in Mv so that it contains a regular fiber of Mv .
Case 2 No simple closed curve in S ∩ Γ2 can be isotoped to be vertical and
in each solid torus component U of Mv \η(Γ2 ), S ∩ ∂U consists of meridians.
In this case, after a small isotopy, S ∩ N (Γ2 ) is a horizontal incompressible
surface. Furthermore, let U be a component of Mv \η(Γ2 ). By Lemma 7.8
S ∩ U consists of meridian disks possibly together with other components that
are obtained by ambient 1–surgery on pairs of meridian disks along a single
arc that joins the two meridian disks. Isotope each such arc out of U , through
N (Γ2 ), avoiding γ , to lie in a component of N (∂Mv ). Then S ∩ U consists of
meridians for each solid torus component of Mv \Γ2 and all ambient 1–surgeries
occur in N (∂Mv ).
Let M̃v be the union of N (Γ2 ) with the solid tori containing the exceptional
fibers of Mv . Then M̃v is a shrunk version of Mv . Note that S ∩ M̃v is a
horizontal incompressible surface. Let T̃ be a component of ∂ M̃v . Then T̃ is
parallel to a decomposing torus T . We replace T by T̃ . We may do so via
an isotopy. We do this for all components of ∂ M̃v . After this process, the
conclusions of the lemma hold.
Case 3 No simple closed curve in S ∩ Γ2 can be isotoped to be vertical and
there is a solid torus component U of Mv \η(Γ2 ) such that S ∩ ∂U does not
consist of meridians.
The argument in this case is given, for instance, in Proposition 1.1 of [2]. For
completeness we povide a sketch of an argument more in line with the ideas used
here. In this case there is a possibly singular annulus A between a component
of S ∩ ∂U and the exceptional fiber f in U . If A is not singular, then A
describes an isotopy of S after which f lies in S . If A is singular, then A
satisfies the hypotheses of Lemma 7.3. Thus f is a core of either V or W .
Geometry & Topology, Volume 8 (2004)
858
Jennifer Schultens
Ie, S is the splitting surface for a Heegaard splitting of M \η(f ). But then
Proposition 7.5 applies to Mv \η(f ) and we may isotope the decomposing tori,
thereby redefining Mv and the edge manifolds for which e is incident to v
slightly, so that after this isotopy, S ∩ Mv is a vertical incompressible surface
and so that an edge manifold becomes the active component.
Lemma 7.10 Suppose that the active component of M = V ∪S W is a vertex
manifold Mv that is a Seifert fibered space with boundary but no exterior
boundary. Suppose further that a fiber f of Mv lies in S . Then one of the
following holds:
(1) We may isotope the decomposing tori, thereby redefining the active component Mv and the edge manifolds for which e is incident to v slightly, so that
after this isotopy, S ∩ Mv is a vertical incompressible surface and so that an
edge manifold becomes the active component.
Or:
(2) S ∩ Mv is pseudohorizontal.
Proof Consider a small regular neighborhood N (f ) of f such that S ∩ N (f )
is a collar A of f . Compress S as much as possible in Mv \η(f ) to obtain an
incompressible surface S ∗ ⊂ M \η(f ). By Haken’s Theorem, each compressing
disk can be chosen to lie entirely on one side of S . Since M = V ∪S W is
strongly irreducible, all compressions must have been performed to one side of
S . It follows that S ∗ lies either in V or in W . There are three options for
S ∗ ∩ Mv :
Case 1 S ∗ ∩ (Mv \η(f )) contains an annulus that is parallel into ∂N (f ).
Note that S ∩ N (f ) is also parallel into ∂N (f ). Thus S ∩ Mv = (S ∗ ∩
(Mv \η(f ))) ∪ (S ∩ N (f )) contains a torus bounding a solid torus in either
V or W . Furthermore, f lies on the boundary of this solid torus and meets a
meridian disk once. After a small isotopy, f is a core of the solid torus. Thus
f is a core of either V or W . But then Proposition 7.5 applies to Mv \η(f )
and we may isotope the decomposing tori, thereby redefining Mv and the edge
manifolds for which e is incident to v slightly, so that after this isotopy, S ∩ Mv
is vertical and so that an edge manifold becomes the active component.
Case 2 S ∗ ∩ (Mv \η(f )) is vertical.
If S ∗ ∩ (Mv \η(f )) is not boundary parallel, then it is essential and can’t be
contained in V or W . This is a contradiction, hence this case does not occur.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
859
Case 3 S ∗ ∩ (Mv \η(f )) is horizontal.
Then (S ∗ ∩ (Mv \η(f ))) ∪ (S ∩ N (f )) is pseudohorizontal, but we must show
that in fact S ∩ Mv = (S ∗ ∩ (Mv \η(f ))) ∪ (S ∩ N (f )).
Since ∂A has two components and since S ∗ ∩ (Mv \η(f )) is separating there are
two, necessarily parallel, components of S ∗ ∩ (Mv \η(f )). Moreover, since A is
parallel into ∂U in both directions, each component of Mv \S ∗ is homeomorphic
to (punctured surface) × (0, 1).
Suppose now that S ∗ lies in, say, V . Denote the component of M \S ∗ that
meets W by Ŵ . Then S defines a Heegaard splitting of Ŵ . Let D be a set of
disks in the interior of Mv that cut Ŵ ∩Mv into a collar of the annuli Ŵ ∩∂Mv .
By Haken’s Theorem, each such disk can be isotoped to intersect S in a single
circle. We may assume that after this isotopy, D still lies in the interior of Mv .
Here S may be reconstructed by performing ambient 1–surgery on S ∗ along arcs
in Ŵ . But this collection of arcs is disjoint from D. Thus all such arcs may be
isotoped into edge manifolds Me such that e is incident to v . Note that an edge
manifold that contains such an arc becomes the active component. Also note
that after this isotopy, the portion of S remaining in Mv is pseudohorizontal.
If a surface is pseudohorizontal, then it is boundary compressible. In particular,
it lives in the active component. Hence the existence of such arcs contradicts
Lemma 6.1. Thus S ∩ Mv = (S ∗ ∩ (Mv \η(f ))) ∪ (S ∩ N (f )).
7.3
Vertex manifold not Seifert fibered
Next we consider the case in which the active component of M = V ∪S W is a
vertex manifold Mv that is homeomorphic to (compact orientable surface) × I .
The strategy here is an adaptation of the argument in the preceeding case.
Though the setup here is much simpler.
Definition 7.11 Let Q be a compact surface with non empty boundary. A
spine of Q is a 1–complex Γ1 with exactly one vertex v that cuts Q into a
regular neighborhood of ∂Q. A 2–complex of the form Γ2 = Γ1 × I is called a
spine of Q × I . We denote the 1–manifold v × I by γ .
The following lemma is another generalization of Lemma 3.11. We will only
be interested in this lemma in the case that N is a vertex manifold of a graph
manifold, but again we state it in very general terms. It too applies to a larger
class of 3–manifolds than just generalized graph manifolds.
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
860
Lemma 7.12 Let M = V ∪S W be a strongly irreducible Heegaard splitting.
Suppose N is a submanifold of M homeomorphic to (compact surface) × I
such that (compact surface) × ∂I ⊂ ∂M . Let Γ2 be a spine of N . Then S may
be isotoped so that the following hold:
(1) S ∩ Γ2 consists of simple closed curves and simple closed curves wedged
together at points in γ .
(2) No closed curve in S ∩ Γ2 bounds a disk in Γ2 \S .
The proof of this lemma is identical to the proof of Lemma 7.7.
Lemma 7.13 Suppose that the active component of M = V ∪S W is a vertex
manifold Mv that is homeomorphic to (compact orientable surface) × I . Then
we may isotope the decomposing annuli, thereby redefining the active component Mv and the edge manifolds for which e is incident to v slightly, so that
after this isotopy, S ∩ Mv is a horizontal incompressible surface and so that an
edge manifold becomes the active component.
Proof Let Γ2 be a spine of Mv and isotope S within Mv so that the conclusions of Lemma 7.12 hold. Then S ∩ Γ2 consists of horizontal curves. Here
S ∩ N (Γ2 ) is a bicollar of S ∩ Γ2 and hence a horizontal incompressible surface.
Let M̃v be N (Γ2 ). Let à be a component of ∂ M̃v \∂Mv . Then à is parallel to
a decomposing annulus A. We replace A by Ã. We may do so via an isotopy.
We do this for all components of ∂ M̃v \∂Mv . After this process, the conclusions
of the lemma hold.
7.4
Edge manifold homeomorphic to (annulus) × I
Now we consider the case in which the active component of M = V ∪S W is an
edge manifold Me homeomorphic to (annulus) × I . This case turns out to be
a direct application of the following theorem of Marty Scharlemann:
Theorem 7.14 Suppose M = V ∪S W is a strongly irreducible Heegaard
splitting and U ⊂ M is a solid torus such that S intersects ∂U in parallel
essential non meridional curves. Then S intersects U in a collection of boundary
parallel annuli and possibly one other component, obtained from one or two
annuli by ambient 1–surgery along an arc parallel to a subarc of ∂U . If the
latter sort of component is in U , then S\U is incompressible in M \U .
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
861
Proof This is Theorem 3.3 in [17].
The following proposition is stated in general terms. Our interest in this theorem will be the case in which A × I is an edge manifold. Since A × I is a solid
torus, this proposition follows directly from Scharlemann’s Theorem (Theorem
7.14).
Proposition 7.15 Suppose A × I is an imbedding of (annulus) × I in the
interior of M with A × {point} essential in M . Further suppose that M =
V ∪S W is a strongly irreducible Heegaard splitting. If both components of
S ∩ (A × ∂I) consist of curves essential in both S and (A × ∂I), then S ∩
(A × I) is isotopic to a collection of incompressible annuli and possibly one
other component, obtained from two annuli by ambient 1–surgery along an arc
parallel to a subarc of A × {point}.
The following definition clarifies the statement of Proposition 7.15.
Definition 7.16 Let A be an annulus and let N = A×I . A spanning annulus
for N is an annulus of the form c × I , for c an essential curve in A. Similarly,
let T be a torus and N = T × I . A spanning annulus for N is an annulus of
the form c × I , for c an essential curve in T .
Each of the incompressible annuli mentioned in Proposition 7.15 is either a
spanning annulus or is parallel into A × ∂I . In Figure 14 we see two isotopic
possibilities. Note that the tube we see on the left hand side is actually “dual”
to the tube we see on the right hand side.
The tube on the left hand side could (after straightening out the picture) be
seen as ambient 1–surgery along an arc parallel to a subarc of {point} × I (a
vertical arc). In this case the the tube on the right hand side would be seen as
ambient 1–surgery along an arc parallel to a subarc of A×{point} (a horizontal
arc).
7.5
Edge manifold homeomorphic to (torus) × I
Next we consider the case in which the active component of M = V ∪S W is
an edge manifold Me homeomorphic to (torus) × I . The theorem proven here
applies in very general contexts. The techniques used are those developed by
Rubinstein and Scharlemann in [16].
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
862
Figure 14: Two isotopic possibilities
The use of these techniques is inspired by Cooper and Scharlemann’s application of the central argument in [16] to the setting of Heegaard splittings of
solvmanifolds in [6, Theorem 4.2]. The arguments in this section are identical
to the argument in [6, Theorem 4.2]. We weaken one of the hypotheses slightly.
In addition, in the setting here, there are no constraints on S ∩ (T × ∂I). In [6]
a constraint arises due to the fact that in a solvmanifold the two components
of T × ∂I are identified. This means that some scenarios that arise in the argument for [6, Theorem 4.2] can be ruled out there. But they can’t be ruled
out here. Thus our conclusions are slightly different.
The argument is rather lengthy. The reader is referred to [6] for a sketch and
to [16] for details.
We recall some fundamentals concerning the Rubinstein–Scharlemann graphic.
Let ΣV be a spine of V and ΣW a spine of W . Then M \(∂− V ∪ΣV ∪∂− W ∪ΣW )
is homeomorphic to S × I . This product foliation is called a sweepout. If S
intersects a product submanifold N = Q × I of M then to each point (s, t) in
I × I we may associate Ss = S × {s} and Qt = Q × {t}. We are interested in
S s ∩ Qt .
After a small isotopy, we may assume that the spines and the sweepout are
in general position with respect to the foliation of Q × I . There is then a 1–
dimensional complex Γ in I ×I called the Rubinstein–Scharlemann graphic. See
Figure 15. At a point (s, t) away from Γ, Ss and Qt are in general position. On
an edge of Γ the surfaces Ss and Qt have a single point of tangency. At a vertex
of Γ there are either two points of tangency or a “birth–death” singularity.
For (s, t) in a region of (I × I)\Γ, Ss ∩ Qt is topologically rigid. If there is a
curve c in Ss ∩ Qt that is essential in Ss but bounds a disk in Qt that lies in V
near c, then we label the region V . Similarly, if there is a curve c′ in Ss ∩ Qt
that is essential in Ss but bounds a disk in Qt that lies in W near c, then we
label the region W .
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
863
Figure 15: The Rubinstein–Scharlemann graphic
We summarize some of the insights from [16]:
Remark 7.17 If a region R is labelled V , then for (s, t) ∈ R the curve c in
Ss ∩ Qt that is essential in Ss and bounds a disk in Qt that lies in V near c
also bounds a disk that lies entirely in V . The equivalent statement holds for
the label W . (This is [16, Lemma 4.3].)
This fact has the following immediate consequences.
Remark 7.18 If a region is labelled both V and W , then M = V ∪S W is
weakly reducible. (This is [16, Corollary 4.4].)
Remark 7.19 If for some s0 ∈ (0, 1) and some t0 , t1 ∈ (0, 1) there is a region
labelled V containing (s0 , t0 ) and a region labelled W containing (s0 , t1 ), then
M = V ∪S W is weakly reducible.
Remark 7.19 is not explicitly stated in [16] but follows immediately from the
definitions and Remark 7.17. Indeed, the labelling in the region containing
(s0 , t0 ) gives an essential disk to one side of Ss0 and the labelling in the region
containing (s0 , t1 ) gives an essential disk to the other side of Ss0 . The boundaries of these disks are contained in Qt0 and Qt1 respectively. As Qt0 and Qt1
are disjoint, the boundaries of these disks are disjoint. Hence M = V ∪S W is
weakly reducible.
More subtle facts are the following:
Remark 7.20 The labels V and W cannot both appear in regions adjacent
along an edge. (This is [16, Corollary 5.5].)
Geometry & Topology, Volume 8 (2004)
864
Jennifer Schultens
Notice that for small s, Ss ∩ (Q × I) is the boundary of a small regular neighborhood of ΣV ∩(Q×I). If the leaves of the foliation of Q×I are essential, then
ΣV can’t miss any such leaf. Similarly for ΣW . Thus we have the following:
Remark 7.21 If the leaves of the foliation of Q × I are essential in M , then
every region of (I × I)\Γ abutting the left edge of I × I is labelled V . Every
region of (I × I)\Γ abutting the right edge of I × I is labelled W .
As a warm up in the application of these principles, we prove the following
lemma:
Lemma 7.22 Suppose that N = Q × I is a product submanifold of M .
Suppose further that M = V ∪S W is a strongly irreducible Heegaard splitting.
Let Γ be the Rubinstein–Scharlemann graphic and suppose there is a vertex
(s0 , t0 ) of Γ with the following properties:
(1) Four regions meet at (s0 , t0 ).
(2) One region is labelled V and one region is labelled W .
Then the following hold:
(A) The other two regions abutting (s0 , t0 ) are unlabelled.
(B) The regions labelled V and W lie opposite each other.
(C) The graph G = Ss0 ∩ Qt0 in Qt0 contains a connected subgraph G̃ with
two vertices v1 , v2 , each of valence 4.
(D) If an edge e of G has both ends on the same vertex v then e ∪ v is an
essential circle in Qt0 .
(E) G̃ is not contractible in Qt0 .
Proof Observations A and B follow directly from Remark 7.20. Recall that
each edge signifies a tangency between Ss0 and Qt0 . If the tangency corresponds
to a maximum or minimum, then the labelling of adjacent regions does not
change as the edge is traversed. So here the tangencies corresponding to edges
must arise from saddle singularities.
It follows that Ss0 ∩ Qt0 consists of a graph G in Qt0 that has two valence four
vertices. In the four regions abutting (s0 , t0 ) the valence four vertices break
apart in four possible combinations. See Figure 16.
Suppose the two valence four vertices of G lie on distinct components G̃1 , G̃2 of
G. As G̃1 , G̃2 break apart as in Figure 16, the curves arising from G̃1 remain
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
865
North
East
West
South
Figure 16: How Ss meets Tt in the four regions North, East, West and South meeting
(s0 , t0 )
disjoint from the curves arising from G̃2 . Furthermore, the curves arising from
G̃1 are identical in the regions North and East and in the regions West and
South. The curves arising from G̃2 are identical in the regions North and West
and in the regions East and South. It follows that one of the components, say
G̃1 , must give rise to the labelling V and the other component, say G̃2 , must
give rise to the labelling W . But then two adjacent regions, say the regions
North and East, are labelled V . A contradiction. Thus C holds.
Denote the vertices by v1 and v2 . Suppose that e has both ends on v1 and that
v1 ∪ e is inessential. Then there will be two adjacent regions, say the regions
North and East, in which this monogon gives rise to a simple closed curve.
Since one of the regions, say North, is labelled, say V , this simple closed curve
must give rise to a labelling V . But then it gives rise to this labelling in both
the region North and in the region East. A contradiction. Hence D holds.
Finally, suppose G̃ is contractible in Qt0 . By D, each edge has endpoints on
distinct vertices. Thus G̃ is as in Figure 17. Hence in the regions North, East,
West and South of (I × I)\Γ we see components of intersection of Ss ∩ Qt in
Qt as in Figure 18.
The curves of intersection pictured are the ones giving rise to the labellings.
Note that if the circles pictured in East and West give rise to labellings, then
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
866
Figure 17: G̃ is contractible
Figure 18: The regions North, East, West and South
they give rise to the same labelling. Hence these regions must both be unlabelled. It follows that North and South are the labelled regions. If the larger
circle pictured in North gives rise to a labelling, then this labelling coincides
with the labelling arising from the circle(s) pictured in South. But this is impossible. Thus the smaller circle pictured in North gives rise to the labelling.
We may assume that it gives rise to the label V .
Let (s, t) ∈ North. Denote the annulus cut out by the two circles in North by
Ã. Denote the disk cut out by the inner circle by D̃ . By Remark 7.17 we may
assume that D̃ ⊂ V . Denote the outer component of ∂ Ã by c and the inner
component by c′ . Since c′ gives rise to a labelling, c′ is essential in Ss . Since
c does not give rise to a labelling, c is inessential in Ss . Let D ⊂ Ss be the
disk bounded by c. We may assume that this disk is disjoint from c ∪ c′ . After
a small isotopy, Ã ∪ D is an essential disk in W . But here ∂(Ã ∪ D) = ∂ D̃ . So
M = V ∪S W is reducible. But this is impossible. Thus G̃ is not contractible
in Qt0 . Hence E holds.
The following proposition gives the required information concerning the intersection of a strongly irreducible Heegaard splitting with an edge manifold
homeomorphic to (torus) × I when this edge manifold is the active component.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
867
Proposition 7.23 Suppose T ×I is an imbedding of (torus)×I in the interior
of M with T × {t} essential in M . Further suppose that M = V ∪S W is a
strongly irreducible Heegaard splitting. If S ∩(T ×∂I) consist of curves essential
in both S and (T ×∂I), then S ∩(T ×I) is characterized by one of the following:
(1) S ∩ (T × I) is isotopic to a collection of incompressible annuli and possibly
one other component, obtained from two such annuli by ambient 1–surgery
along an arc parallel to a subarc of T × {point}.
(2) There is a pair of simple closed curves c, c′ ⊂ T such that c ∩ c′ consists of
a single point p ∈ T and V ∩ (T × I) is a collar of (c × {0}) ∪ (p × I) ∪ (c′ × {1}).
Proof Consider the regions of (I × I)\Γ abutting I × {0}. Since S ∩ (T × ∂I)
consist of curves essential in both S and (T × ∂I) there must be an unlabelled
such region. We call this region R0 . Similarly, there must be an unlabelled
region, which we call R1 , that abuts I × {1}. Because of the conditions on
S ∩ (T × ∂I) we may assume that for (s, t) in R0 or R1 , Ss ∩ Tt consists of
curves essential in both Ss and Tt . We are interested in monotone paths from
R0 to R1 .
Our argument breaks into two cases: Either there is a path in I × I from R0
to R1 that avoids labelled regions or there is no such path.
Case 1 There is a path beginning in R0 and ending in R1 that traverses only
unlabelled regions and edges between such regions. See Figure 19.
Figure 19: A monotone path through unlabelled regions
It follows from Remark 7.19 that there is a monotone path α beginning in R0
and ending in R1 that traverses only unlabelled regions and edges between such
regions.
Consider the effect of traversing an edge e from one unlabelled region of (I ×
I)\Γ to another. Since compression bodies do not contain essential surfaces,
Geometry & Topology, Volume 8 (2004)
868
Jennifer Schultens
Ss ∩ Tt must contain curves essential in Ss . For (s, t) in an unlabelled region,
such curves must also be essential in Tt . We may thus refer to curves inessential
in both Ss and Tt simply as inessential and to curves essential in both Ss and
Tt simply as essential.
As α crosses e, the components of intersection of Ss ∩ Tt change in the same
way that level curves change, as we rise from being below to being above a
maximum, minimum or saddle point. If the point of tangency corresponding to
e corresponds to a maximum or minimum, then an inessential curve appears
or disappears. If the point of tangency corresponds to a saddle singularity,
then either one curve is banded to itself or two curves are banded together.
Note that Ss ∩ Tt is separating in Tt , hence the essential curves come in pairs
parallel in Tt . Thus if a curve is banded to itself, then any resulting essential
curves are parallel, in Tt , to other essential curves. If two essential curves are
banded to each other, then an inessential curve results, but essential curves must
remain. In both cases, the slope, in Tt , of the essential curves is unaffected, as
t increases.
Since α is monotone, we obtain a function f : I → I by requiring f (t) to be the
value s such that (s, t) is in the image of α. We may now isotope S ∩(T ×I) by
isotoping S ∩ (T × {t}) to Sf (t) ∩ Tt for each t ∈ I . Thus α describes an isotopy
of S . We will assume in the following that this isotopy has been performed.
The preceding paragraph tells us how S ∩ (T × {t}) changes as t increases. We
must reconstruct S ∩ (T × I) from these level sets.
As α traverses an unlabelled region, the curves of intersection S ∩ Tt sweep
out annuli. As α crosses an edge corresponding to a tangency corresponding
to a maximum or minimum an inessential curve appears or disappears. If
such a curve appears, this indicates the appearance of a disk. If such a curve
disappears, this indicates that an inessential curve is capped off by a disk. Note
that since there are no disks for t = 0 or for t = 1, these disks merely cap off
inessential curves. Converserly, inessential curves are always capped off by such
disks.
Suppose that α crosses an edge corresponding to a tangency corresponding to
a saddle point at (s0 , t0 ). If an inessential curve is banded to itself, then there
are two possibilities:
(1) Two inessential curves result. These curves are nested in Tt0 +ǫ . However, one of these inessential curves lies in the subdisk of S bounded by the
other. Hence the appearance of the new inessential curve does not indicate
the appearance of another disk. The appearance of the new inessential curve
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
869
merely indicates a saddle singularity, with respect to projection onto t, in the
imbedding of a single disk. Thus we may ignore this phenomenon.
(2) Two essential curves result. Since the inessential curve bounds a disk in
S , the components of S ∩ Tt affected piece together to form an annulus that is
parallel into Tt0 +ǫ .
If an essential curve is banded to itself, then either one essential curve or one
inessential and one essential curve result. But the first possibility can’t occur
here, because here S ∩ Tt is separating for all t. Thus one inessential and one
essential curve result. The inessential curve bounds a disk in S . Hence the
appearance of the new inessential curve merely indicates a saddle singularity,
with respect to projection onto t, in the imbedding of an essential annulus.
Thus we may again ignore this phenomenon.
Finally, suppose that a pair of essential curves disappears as two essential curves
are banded to each other to produce an inessential curve. The inessential curve
bounds a disk in S . Thus the components of S ∩ Tt affected piece together to
form an annulus that is parallel into Tt0 −ǫ .
Since there are no inessential curves for t = 0 or t = 1, we see that, after
further isotopies, S ∩ (T × I) consists of spanning annuli and annuli parallel
into T × ∂I .
Case 2 There is no path beginning in R0 and ending in R1 that traverses
only unlabelled regions and edges between such regions. See Figure 20.
Figure 20: A monotone path through unlabelled regions
In this case the regions of (I × I)\Γ labelled V extending from the left edge of
I × I must meet the regions of (I × I)\Γ labelled W extending from the right
edge of I × I . By Remark 7.20 this can only happen at a vertex. Thus Lemma
7.22 applies.
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
870
We consider G̃ as in C) of Lemma 7.22. Denote the vertices by v1 , v2 and the
edges by e1 , e2 , e3 , e4 . Suppose first that one edge, say e1 has both ends on v1 .
Then since G̃ is connected, and since v1 and v2 have valence four, there must
be exactly two edges e2 , e3 , each with one end on v1 and one end on v2 . It
follows that e4 must have both ends on v2 . By Lemma 7.22 both v1 ∪ e1 and
v2 ∪ e4 are essential. Since these curves lie on a torus, they are parallel. The
other possibility is that all edges have one end on v1 and one end on v2 .
Subcase 2.1 e1 has both ends on v1 , e4 has both ends on v2 and e2 and e3
are parallel.
Here G̃ is as in Figure 21.
Figure 21: G̃ contains a bigon
This implies that as v0 is crossed going from one unlabelled region to another,
S ∩ Tt changes as in Figure 22.
Figure 22: Adding a 1–handle
But this corresponds to the addition of a 1–handle. The core of the 1–handle
can be isotoped into Tt0 . Since there are essential disks for both V and W cut
out by G̃, S\(T × [t0 − ǫ, t0 + ǫ]) is incompressible in M \(T × [t0 − ǫ, t0 + ǫ]).
In particular, we may assume that S ∩ (T × {t0 − ǫ, t0 + ǫ}) consists of curves
essential in S and in Tt0 ±ǫ . This implies that S ∩ (T × [t0 − ǫ, t0 + ǫ]) is as
described in option (1).
To see that option (1) holds for all of S ∩ (T × I), note that S ∩ ((T × I)\(T ×
[t0 −ǫ, t0 +ǫ])) consists of incompressible annuli. Spanning annuli merely extend
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
871
components of S ∩ (T × [t0 − ǫ, t0 + ǫ]). Annuli parallel into Tt0 ±ǫ create
annuli parallel into T × ∂I unless they meet the compressible component of
S ∩ (T × [t0 − ǫ, t0 + ǫ]).
The question is thus merely whether the ends of the compressible component
of S ∩ (T × [t0 − ǫ, t0 + ǫ]) meet spanning annuli or annuli parallel into Tt0 ±ǫ .
There are a number of possibilities. But all possibilities are either as described
in option (1) or lead to a contradiction, by implying that S is stabilized. Some
options are pictured in Figure 23, Figure 24, Figure 25.
Figure 23: Compressible component meets annulus parallel into Tt0 ±ǫ
Figure 24: Another stabilized possibility
Subcase 2.2 e1 has both ends on v1 , e4 has both ends on v2 and e2 and e3
are not parallel.
Here G̃ is as in Figure 26.
This implies that as v0 is crossed going from one unlabelled region to another,
S ∩ Tt changes as in Figure 27.
Thus S ∩ T × [t0 − ǫ, t0 + ǫ] consists of spanning annuli together with one
component that is obtained as follows: an annulus that is parallel into Tt0 ±ǫ
is tubed to a spanning annulus. This is precisely the type of compressible
Geometry & Topology, Volume 8 (2004)
Jennifer Schultens
872
Figure 25: Two isotopic possibilities
Figure 26: G̃ contains a triangle (dashed arc included along with G̃ to indicate that a
bicoloring, as required, is possible)
Figure 27: Adding an “essential” 1–handle
component that arises in Case 2.1 if an annulus parallel into Tt0 −ǫ is attached
to one end of the compressible component and one end of a spanning annulus.
Thus as in Subcase 2.1, all unstabilized configurations involving S ∩ T × [t0 −
ǫ, t0 + ǫ] are as described in option (1).
Subcase 2.3 All edges have one end on v1 and the other on v2 .
Note that S is separating. This induces a bicoloring of Tt0 \S . Such a bicoloring
is not possible if two edges are parallel. This forces G̃ to be as in Figure 28.
This implies that as v0 is crossed going from one unlabelled region to another,
S ∩ Tt changes as in Figure 29.
In particular, S ∩ T × [t0 − ǫ, t0 + ǫ] is as in option (2). Again, S ∩ ((T × I)\(T ×
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
873
Figure 28: G̃ contains a square
Figure 29: Another way of adding a 1–handle
[t0 −ǫ, t0 +ǫ])) consists of incompressible annuli. If an end of S ∩T ×[t0 −ǫ, t0 +ǫ]
meets an annulus that is not isotopic to a spanning annulus, then both ends
meet this annulus. But this would imply that S ∩ (T × I) is disconnected.
Hence the ends of S ∩ T × [t0 − ǫ, t0 + ǫ] meet spanning annuli and S ∩ (T × I)
is as in option (2). See Figure 29.
8
Fitting the pieces together
We here prove the main theorem. As it turns out, the hard work is already
done. It remains to fit the results together.
Recall that graph manifolds with empty characteristic submanifolds are Seifert
fibered spaces. The corresponding result for Seifert fibered spaces was obtained
by Y. Moriah and the author in [13] together with [21] and [22]. Generalized graph manifolds with empty characteristic submanifolds are either Seifert
fibered spaces or product manifolds of the form (surface) × I . The corresponding result for such manifolds was obtained by M. Scharlemann and A.
Thomspon in [19].
Proof of Theorem 1.1 If E = ∅, then S is either pseudohorizontal or pseudovertical, by the main theorems of [19], [13], [21] and [22]. Thus we may
assume in what follows that E =
6 ∅.
Geometry & Topology, Volume 8 (2004)
874
Jennifer Schultens
Let T be the collection of decomposing annuli and tori. Let N be a component
of M \T . By Lemma 6.1, S ∩ N is incompressible unless N is the active
component. Thus if N is not the active component, then S ∩ N is as required.
Specifically, if N is a vertex manifold, S ∩ N is either horizontal or vertical.
And if N is an edge manifold, then S ∩ N consists of incompressible annuli.
If N is the active component, we must consider the possibilities: If N is a
vertex manifold that is Seifert fibered and has non empty external boundary
then Proposition 7.5 applies. If N is a vertex manifold that is a product, then
Lemma 7.13 applies. Thus in these two cases, we may rechoose the decomposing
annuli or tori so that an edge manifold becomes the active component. Similarly,
if N is vertex manifold that is a Seifert fibered space with no external boundary,
then either S ∩ N is pseudohorizontal, or we may rechoose the decomposing
tori so that an edge manifold becomes the active component.
Finally, if N is an edge manifold, then S ∩ N is as required by Proposition 7.15
and Proposition 7.23.
Now Theorem 1.2 and Theorem 1.3 follow from Theorem 1.1 along with the
constructions of untelescoping and amalgamation. Recall Theorem 3.10 which
states that, given an irreducible Heegaard splitting, the amalgamation of the
weak reduction of this Heegaard splitting yields the original Heegaard splitting.
Note that the amalgamation of strongly irreducible Heegaard splittings of generalized graph manifolds interferes with the structure of the Heegaard splittings
on the generalized graph manifolds that are being amalgamated. For this reason, we can’t prove 1.1 without the hypothesis of strong irreducibility. Recall
Example 5.6, which illustrates the obstruction.
Proof of Theorem 1.2 Consider the irreducible Heegaard splitting M =
V ∪S W . Let M = (V1 ∪S1 W1 ) ∪F1 · · · ∪Fn−1 (Vn ∪Sn Wn ) be a weak reduction of M = V ∪S W . Denote the decomposing tori by T . By Lemma 4.2,
∪i Fi can be isotoped so that for each vertex manifold Mv of M , Fi ∩ Mv is
either horizontal or vertical and so that for each edge manifold Me , Fi ∩ Me
consists of incompressible tori and essential annuli. Hence, cutting M along
∪i Fi consists of generalized totally orientable graph manifolds.
Since M = (V1 ∪S1 W1 ) ∪F1 · · · ∪Fn−1 (Vn ∪Sn Wn ) is a weak reduction of
M = V ∪S W , each Mi = Vi ∪Si Wi is strongly irreducible. Thus Mi = Vi ∪Si Wi
is a strongly irreducible Heegaard splitting of a generalized graph manifold
and Theorem 1.1 applies. The result now follows by juxtaposing the strongly
irreducible Heegaard splittings of the generalized graph manifolds.
Geometry & Topology, Volume 8 (2004)
Heegaard splittings of graph manifolds
875
The Euler characteristic calculation follows directly from the definition of a weak
reduction of a Heegaard splitting along with the fact that here χ(∂M ) = 0.
Proof of Theorem 1.3 This follows immediately from Theorems 3.10 and
1.2.
Note that a horizontal (or a pseudohorizontal, respectively) surface in a vertex
manifold corresponds to a foliation of that manifold as a surface bundle over the
circle (or to a foliation of that manifold minus a vertical solid torus as a surface
bundle over the circle). If the splitting surface of a Heegaard splitting can
be isotoped so as to be horizontal or pseudohorizontal in two adjacent vertex
manifolds, then the glueing data must satisfy certain restrictions. It is thus often
the case that the splitting surface of a strongly irreducible Heegaard splitting
of a graph manifold can’t be isotoped to be horizontal or pseudohorizontal in
adjacent vertex manifolds. Consequently, the canonical example of a Heegaard
splitting of a graph manifold would seem to be a Heegaard splitting obtained by
taking Heegaard splittings of the vertex manifolds with pseudovertical splitting
surfaces and amalgamating these to obtain a Heegaard splitting of the graph
manifold.
References
[1] M Boileau, J-P Otal, Groupes des difféotopies de certaines variétés de Seifert,
C.R. Acad. Sci. Paris 303-I (1986) 19-22, MR87g:57022
[2] M Boileau, J-P Otal, Scindements de Heegaard et groupes des homotopies
des petites varietes de Seifert, Invent. Math. 106 (1991) 85–107
[3] F Bonahon, J-P Otal, Scindements de Heegaard des espaces lenticulaires,
Ann. Scient. Ec. Norm. Sup. 16 (1983) 451-466
[4] G Burde, H Zieschang, Knots, de Gruyter Studies in Mathematics, 5. Walter
de Gruyter & Co. Berlin (1985)
[5] A Casson, C Gordon, Reducing Heegaard splittings, Topology and its Applications 27 (1987) 275–283
[6] D Cooper, M Scharlemann, The structure of a solvmanifold’s Heegaard splittings from: “Proceedings of 6th Gökova Geometry–Topology Conference”, Turkish J. Math. 23 (1999) 1–18, arXiv:math.GT/9803157
[7] D B A Epstein, Periodic flows on 3–manifolds Ann. Math. 95 (1972) 68–82
[8] D Gabai, Foliations and the topology of 3–manifolds III, J. Differential Geom.
26 (1987) 479–536
Geometry & Topology, Volume 8 (2004)
876
Jennifer Schultens
[9] J Hempel, 3–manifolds, Annals of Mathematics Studies 86, Princeton University Press, Princeton, NJ (1976)
[10] W Jaco, Lectures on Three-Manifold Topology, CBMS Lecture Notes, No. 43,
Amer. Math. Soc. Providence, RI (1980) MR81k:57009
[11] W B Raymond Lickorish, An introduction to knot theory, Graduate Texts in
Mathematics 175, Springer–Verlag, New York (1997)
[12] J Milnor, Lectures on the h–cobordism theorem, notes by L Siebenmann and L
Sondow, Princeton Mathematical Notes, Princeton University Press (1965)
[13] Y Moriah, J Schultens, Irreducible Heegaard splittings of Seifert fibered spaces
are horizontal or vertical, Topology 37 (1998) 1089–1112
[14] J. Nielsen, Abbildungsklassen endlicher Ordnung, Acta Mathematica 75 (1943)
23–115
[15] Dale Rolfsen, Knots and links, Corrected reprint of the 1976 original, Mathematics Lecture Series, 7, Publish or Perish, Inc. Houston, TX (1990)
[16] J Hyam Rubinstein, M Scharlemann, Comparing Heegaard splittings of
non-Haken 3–manifolds Topology 35 (1996) 1005–1026
[17] M Scharlemann, Local detection of strongly irreducible Heegaard splittings
Topology and its Applications 90 (1998) 135–147
[18] M Scharlemann, A Thompson, Thin position for 3–manifolds AMS Contemporary Math. 164 (1994) 231–238
[19] M Scharlemann, A Thompson, Heegaard splittings of (surf ace) × I are
standard, Math. Ann. 295 (1993) 549–564
[20] J Schultens, The Classification of Heegaard splittings for (closed orientable
surf ace) × S 1 , London Math. Soc. 67 (1993) 425–448
[21] J Schultens, Heegaard splittings of Seifert fibered spaces with boundary, Trans.
Amer. Math. Soc. 347 (1995) 2533–2552
[22] J Schultens, Weakly reducible Heegaard splittings of Seifert manifolds are vertical, Topology Appl. 100 (2000) 219–222
[23] J Schultens, Additivity of Tunnel Number for Small Knots, Comment. Math.
Helv. 75 (2000) 353–367
[24] J Schultens, R Weidmann, On the geometric and algebraic rank of graph
manifolds, preprint
[25] William P Thurston, Silvio Levy, editor, Three-dimensional geometry and
topology. Vol. 1 Princeton Mathematical Series, 35. Princeton University Press,
Princeton, NJ (1997) ISBN: 0-691-08304-5
[26] R Weidmann, Some 3–manifolds with 2–generated fundamental group, to appear in Arch. Math.
Geometry & Topology, Volume 8 (2004)